The struggle to catch up in software
March 19, 1979
Software is clearly the throttle that keeps sales of the computer-on-a-chip from growing even faster than they already are. Writing the programs that tell the computer what to do is still not only the biggest problem for users but also the most costly. And until recently, microprocessor users were not getting all that much software help from the manufacturers. Now that is changing, and manufacturers are beginning to invest large amounts of money and people in the job of making it easier and cheaper to use their revolutionary devices.
Fast-moving industry trends make that job a formidable one. Microprocessor hardware continues to plunge in price, but the cost of writing the operating and application programs is moving in the opposite direction just as fast. Soaring labor costs, accentuated by a shortage of experienced programmers, have more than doubled the cost of some software. Five years ago software accounted for as little as 5% of the cost of developing a microprocessor system. Now it can be as much as 80%. As Dennis R. Allison, a Menlo Park (Calif.) programming consultant, puts it: "Many potential users face the prospect of spending $150,000 on software to run a device that costs $10. The scale here is disturbing."
To complicate matters even more, the microprocessor -- unlike larger computers -- is sold mainly to computer novices. "We're selling microcomputers to gear, lever, and pulley people," says William B. Sweet, marketing head for microcomputer systems at National Semiconductor Corp. "Software for them is nothing but an intimidating bottleneck."
But the seven-year-old microprocessor industry is following the same route that the minicomputer industry took a decade ago. It was a long time before mini makers were able to provide sophisticated software packages for their low-priced "black boxes," observers recall. Microprocessor makers are now addressing the software bottleneck with more urgency and resources than ever before. And they are hustling to bring out an array of software paraphernalia and development aids for users.
Software and system-support products, for example, accounted for nearly 75% of the total cost of developing Intel Corp.'s new high-powered, 16-bit microprocessor. And Motorola Inc. spent half of the development money for its new 16-bit processor on software, compared with just 20% of the money used to develop its older, medium-powered, 8-bit microprocessor.
For users, the biggest payoff in this growing emphasis on software by manufacturers is the introduction of so-called high-level languages that are easier and faster to use in writing programs. Earlier microprocessor languages are both time-consuming and cumbersome. Writing a long program in one of the older languages -- such as the 65,000-character effort that Tektromix Inc. completed recently for one of its new "smart" graphics terminals -- takes about eight man-years of programming at a cost approaching $300,000.
Switching to a high-level language increases the productivity of a programmer dramatically, because a single line of high-level language can do a job that takes about five lines of an older language. The cost savings are huge, because the average programmer can write only about 10 good lines of program code a day.
The growing role of Pascal
The problem in the past was that memory was too expensive for high-level languages, which require much larger chunks of memory than the older languages. "But we knew as the microprocessor chips got more complex and memory got cheaper that people would start embracing high-level languages," says Lyman C. Hevle, microsystems marketing manager at Motorola Inc. AS a result, more and more manufacturers are now offering their customers the tools for such high-level languages as Cobol, Basic, Fortran, and -- perhaps even more important -- the much touted "structured" languages such as Pascal.
Backers of such new languages as Pascal are downright rabid in their enthusiasm. They claim that these languages can make programming as much as 10 times faster and can cut the cost of software 30% to 75%. Both Motorola and Texas Instruments Inc. have jumped on the Pascal bandwagon, making the language a major part of their microprocessor strategy.
TI is hoping that Pascal and its other software capabilities will help it to catch up with microprocessor leaders such as Intel. The semiconductor giant from Texas figures that it has a leg up on the overall software problem because it has spent the last three years developing a software base. TI was one of the first semiconductor makers to introduce a 16-bit microprocessor, although the device has been slow to catch on.
One key advantage for the customer in switching to a high-level language is what the industry calls portability. When software is written in a high-level language such as Pascal, the programs can readily be switched from one microprocessor to another with only minor changes. In the past, users were essentially locked into one microprocessor design, since each processor had its own unique machine language. If a user wanted to change to another make or model, it meant dumping all of his investment in software. "One of our customers' biggest software hangups has been that they would spend 10 to 15 man-years on software development and three years later -- in order to upgrade hardware -- have to go back and start all over again on software," comments Dennis E. Starbuck, an applications engineering manager at Rockwell International Corp.
Some semiconductor makers, such as Intel, still adamantly resist the idea of "chip-independent languages," but the advantages seem overwhelming for many customers. Kenneth L. Bowles, professor of computer science at the University of California at San Diego, already has demonstrated that, with only a small amount of additional software, the same operating programs written in Pascal can run on half-a-dozen different microprocessors.
Another assault on software costs now picking up speed is the introduction of standardized software by the microprocessor makers. Motorola, for example, this year will introduce standard software packages stored in permanent semiconductor memories called ROMs, a product that the industry calls firmware. The plug-in software will be written in blocks that can be strung together in any order. The company also plans to announce several standard microprocessors -- firmware packaged with a processor. "Firmware gets software on the hardware learning curve, lowering its price and eventually making it uneconomical for the customer to write his own software," says Murray A. Goldman, microcomputer operations manager at Motorola.
Standard software will be a difficult concept to sell, despite its significant cost savings. But Motorola is not the only microprocessor maker moving in this direction. Texas Instruments, in fact, is out in front with several standard modules already on sale, having successfully cut its teeth by marketing firmware for its programmable handheld calculator. This year, TI will bring out several more of its "component" modules for building application programs, including software for communication protocols, file management, and process control. Says one delighted early user of the standard modules: "It's like having a spare-parts bin."
For TI, turning out blocks of software in semiconductor memory is a coner-stone of its strategy for growth in distributed computing, one of the company's major market thrusts in the 1980s. "Conceptually, it's like what we do with hardware chip sets," says W. Kenneth Wickham, TI's manager for microprocessor software. "We prefabricate certain primary functions that are used frequently and sell them just like ICs that can be hooked together to form a circuit."
A lock on the market
Easily the most important software design aid now being offered is the microprocessor development system, which vastly simplifies the job of writing application programs. Such a system, which usually consists of a processor, keyboard, video display, and printer, does its own diagnostics and takes the user step by step through the production of a software program. National Semiconductor introduced its Starplex system last fall "on the assumption that the user is a real dummy when it comes to microprocessors and programming," says National's Sweet. "With Starplex, the user doesn't even have to read an operating manual. It's almost as easy as filling in the blanks."
While that claim seems a bit exaggerated, the development systems are beginning to sell in large numbers. The leading beneficiary, by far, is Intel, which is said to be shipping about 400 of its $15,000 systems a month. By the end of 1979, the number of installed Intel development systems should be close to 16,000 -- a $250 million customer investment -- "making Intel's lock on the microcomputer market even more formidable than it is today," predicts Benjamin M. Rosen, a veteran industry watcher at Morgan Stanley & Co.
In many cases, the solution to the critical shortage of programmers has been to teach engineers how to program. Integrated Computers System Inc. has run 10,000 engineers and managers through its week-long courses on programming microprocessors in the last five years. "But you can't just hire an average programmer off the street," says Jarad A. Anderson, president of Two Pi Corp., a major user of microprocessors, because "almost all microprocessor programming requires some knowledge of the hardware that is going to use the microprocessor."
The microprocessor has had one factor in its favor regarding software. In a large percentage of the applications -- games and appliance controllers, for example -- software has been a one-time cost amortized over thousands, sometimes even millions, of units. Mattel Toys Inc.'s Electronics Div. now uses microprocessors in a dozen of its handheld games, and we're doing such tonnage of product that the high cost of software is no problem," says Jeffrey A. Rochliss, division president.
But the new generation of high-power microprocessors is changing that picture. The 16-bit devices, at least for now, are tending to go into lower-volume products, even into "one-of-a-kind" control systems, where the software costs cannot be amortized over a large number of end products. And despite the greatly increased spending by vendors on software and support, makers are not getting high marks for their new effort in all cases. The operating manuals on the new 16-bit devices are "nightmares of cryptic computerese," declares Wayne C. Kahn, president of National Microcomputer Inc., a custom systems house.
The microprocessor makers know they have a problem here. "With the 16-bit microprocessor, the user is going to demand the ultimate in ease of programming," acknowledges Robert E. Anslow, director of microelectronic business development at Rockwell. But at this point the industry can only guess at what the "ultimate" will be. there is talk of software factories that would efficiently and cheaply grind out the miles of code needed to run microcomputers, and of facilities that would use low-scale labor to write standard modules of programs. But such solutions, if they come, are many years away.
"Software undeniably is a limitation to the use of microprocessors," acknowledges Anderson of Two Pi. "But if microprocessors are indeed bringing the second industrial revolution, and the first one took 100 years, we've got to at least be willing to give this one 25."
GRAPHIC: Picture, consultant Bowles: A language that allows a program to run on many different processors. James Caccavo/Empire
Copyright 1979 McGraw-Hill, Inc.